Content Coach

AI Content Coach — Setup & Run

Trigger

User types /content-coach or /content-coach setup

What This Does

Walks the user through setting up the AI Content Coach dashboard — a local intelligence tool that scrapes their Instagram posts and competitors, transcribes reels, and gives them a Claude-powered chat interface for content strategy.

STEP 1 — Welcome

Tell the user:

"Let's get your AI Content Coach dashboard running. I'll walk you through every step — this takes about 10 minutes. You'll need:

• Node.js installed
• A free Apify account (for Instagram scraping)
• Python 3 (for video transcription)
• Your Instagram handle + 3–5 competitors to track"

Ask: "Ready? Where did you clone the repo? (Paste the full folder path)"

Save this as REPO_DIR. All commands below run from this directory.

STEP 2 — Check Prerequisites

Run all checks in parallel. For each failure, provide the exact install command.

Node.js (v18+)

node --version

If missing or < v18: brew install node (Mac) or direct them to nodejs.org

Python 3

python3 --version

If missing: brew install python3 (Mac) or python.org

ffmpeg (required by yt-dlp and Whisper)

ffmpeg -version

If missing: brew install ffmpeg

yt-dlp (downloads Instagram reels)

yt-dlp --version

If missing:

pip3 install yt-dlp

After installing, find the binary path:

which yt-dlp || python3 -m site --user-base

Whisper (transcribes reels to text)

python3 -c "import whisper; print('Whisper OK')"

If missing:

pip3 install openai-whisper

Note: Whisper downloads a ~150MB model on first run. The base model is fast enough for this use case.

After all prerequisites pass, tell the user which ones you installed and confirm everything is green.

STEP 3 — Claude Auth Method

Ask:

"How are you running Claude Code?

A) I have an Anthropic API key (from console.anthropic.com) B) I use a Claude.ai subscription (the $20/month plan)

Type A or B."

If A (API key): Ask them to paste their API key. Add to .env:

ANTHROPIC_API_KEY=sk-ant-...

Tell them: "The server will use your API key directly — no extra setup needed."

If B (Claude subscription): Tell them: "The server routes through your installed Claude Code — no API key needed. The chat just works." Do NOT add anything to .env for auth.

STEP 4 — Instagram Handle

Ask:

"What's your Instagram handle? (no @ symbol, just the username)"

Example: tenfoldmarc

Save as YOUR_HANDLE.

STEP 5 — Competitor Handles

Ask:

"Give me 3–5 competitor Instagram handles you want to spy on. These are accounts in your niche you want to benchmark against.

List them one per line or comma-separated. (no @ symbols)"

Example:

chase.h.ai
noevarner.ai
mattganzak
agentic.james

Save as COMPETITORS list.

STEP 6 — Apify Token

Ask:

"Now we need your Apify token — this is what scrapes Instagram for you.

1. Go to apify.com and create a free account
2. In your Apify dashboard → Settings → API tokens
3. Copy your Personal API token
4. Paste it here"

Add to .env:

APIFY_TOKEN=apify_api_...

Tell them: "Apify charges ~$0.002–0.006 per profile scraped. Pulling 25 posts from 5 accounts costs about $0.03."

STEP 7 — Write Config Files

Create config.json in REPO_DIR:

{
  "yourHandle": "YOUR_HANDLE",
  "yourName": "Their Name",
  "postsPerAccount": 25,
  "competitors": [
    { "handle": "competitor1", "name": "Competitor 1 Name", "pattern": "" },
    { "handle": "competitor2", "name": "Competitor 2 Name", "pattern": "" }
  ]
}

Fill in their actual handles. For "name" fields, use the handle as the name for now (they can update it). For "pattern" — leave empty string. It gets populated by Claude's analysis after the first data fetch.

Confirm .env has:

APIFY_TOKEN=apify_api_...
# Only if they chose option A:
ANTHROPIC_API_KEY=sk-ant-...

STEP 8 — Install Node Dependencies

cd REPO_DIR && npm install

Confirm it completed without errors.

STEP 9 — Find Tool Paths

Run these to find exact binary paths (needed for transcription):

which yt-dlp || python3 -m site --user-base
which whisper || python3 -c "import whisper; import os; print(os.path.dirname(whisper.__file__))"
which ffmpeg

These paths get written into config.json automatically by the next step. Tell the user: "Found your tools — locking in the paths so the scraper always knows where they are."

Update config.json to add:

{
  "tools": {
    "ytdlp": "/path/to/yt-dlp",
    "whisper": "/path/to/whisper",
    "ffmpeg": "/path/to/ffmpeg",
    "whisperModel": "base"
  }
}

STEP 10 — Initial Data Fetch

Tell the user: "Now scraping your last 25 posts + all competitors. This takes 2–4 minutes. Don't close the terminal."

cd REPO_DIR && node fetch-data.js

Watch for output. If it errors:

• APIFY_TOKEN not found → check .env file
• Apify run failed → check token is valid at apify.com
• yt-dlp errors → some reels may be private, that's OK, they get skipped

When complete, confirm: "X posts saved from your account, Y posts saved from Z competitors. Transcriptions complete."

STEP 11 — Start the Server

cd REPO_DIR && node server.js

Then open the browser:

open http://localhost:3003

Tell the user:

"Your AI Content Coach is live at http://localhost:3003

Every time you want fresh data, run: node fetch-data.js Every time you want to open the dashboard, run: node server.js

The AI chat in the dashboard is trained on ALL your saved posts — the more you run fetch-data.js, the smarter it gets."

Error Handling

• Port 3003 in use: Tell them to run node server.js -- --port 3004 or kill the process using lsof -i :3003
• Instagram scrape returns 0 posts: Some accounts have rate limiting. Tell them to wait 30 min and retry.
• Whisper model download: First run downloads ~150MB. Tell them it's a one-time thing.
• Private Instagram account: yt-dlp needs Instagram cookies for private accounts. Skip transcription for private accounts, metadata-only still works via Apify.

Ongoing Commands

After setup, when user types /content-coach:

• If dashboard isn't running: start it (node server.js && open http://localhost:3003)
• If they want fresh data: run node fetch-data.js then restart server
• If they want to add a competitor: update config.json then re-run fetch